tio

ez

h i g h l i g h t s

This paper presents the Semi-Empirical Urban SEUS is a simple model to estimate air pollut The performance of SEUS is evaluated in a str Estimated and observed NOx and NO2 hourly

rate, in

Street canyonModelling

in other input variables. The model estimates NO2 concentrations using a simple photochemistryscheme. SEUS is applied to estimate NOx and NO2 hourly concentrations in an irregular and busy street

streets are strongly affected by emissions taking place inside thestreet itself. However, the concentration level and the distributionof air pollution inside the street are governed by surrounding

ily affect the windthe street (Brittermple is the streetn inside the streetcular to the street

canyon a spiral or helical type ow develops within the canyon.Within the vortex ow relatively clean air from rooftop height isdrawn down at the windward face of the street, across the road atstreet level, in the reverse of the wind direction at rooftop, bringingpollutants in the road to leeward face of the canyon (Britter andHanna, 2003). The concentration inside the urban street canyoncan be considered as the result of two contributions, one fromemissions from local trafc in the street itself and one from back-ground pollution entering the street canyon from above roof level.

* Corresponding author.E-mail addresses: lvenegas@bertel.com.ar, lvenegas@fra.utn.edu.ar (L.

E. Venegas), nmazzeo@fra.utn.edu.ar (N.A. Mazzeo), ymariana@gmail.com (M.

Contents lists availab

Atmospheric E

journal homepage: www.else

Atmospheric Environment 87 (2014) 77e86C. Dezzutti).1. Introduction

Trafcked streets are air pollution spots in the urban areas. Anurban street canyon is characterised by the presence of buildings onboth sides of the street. Naturally, the pollutant levels in the urban

physical conditions. The physical conditions heavspeed and especially the wind direction insideand Hanna, 2003; Oke, 1996). The classical exavortex ow that governs the pollutant distributiocanyon. When wind ows are close to perpendisignicant discrepancies have been found between the results of the proposed model and that of awidely used operational dispersion model (OSPM), both using the same input information.

2014 Elsevier Ltd. All rights reserved.Buenos Aires canyon in the city of Buenos Aires. The statistical evaluation of results shows that there is a goodagreement between estimated and observed hourly concentrations (e.g. fractional bias are 10.3% forNOx and 7.8% for NO2). The agreement between the estimated and observed values has also beenanalysed in terms of its dependence onwind speed and direction. The model shows a better performancefor wind speeds >2 m s1 than for lower wind speeds and for leeward situations than for others. No SEUS model has been reasonable accu

a r t i c l e i n f o

Article history:Received 7 August 2013Received in revised form19 November 2013Accepted 3 January 2014Available online 14 January 2014

Keywords:Urban pollution1352-2310/$ e see front matter 2014 Elsevier Ltd.http://dx.doi.org/10.1016/j.atmosenv.2014.01.005Street (SEUS) model.ant concentrations in street canyons.eet canyon in the city of Buenos Aires.concentrations show good agreement.spite of its simplicity.

a b s t r a c t

This paper introduces the Semi-Empirical Urban Street (SEUS) model. SEUS is a simple mathematicalmodel based on the scaling of air pollution concentration inside street canyons employing the emissionrate, the width of the canyon, the dispersive velocity scale and the background concentration. Dispersivevelocity scale depends on turbulent motions related to wind and trafc. The parameterisations of theseturbulent motions include two dimensionless empirical parameters. Functional forms of these param-eters have been obtained from full scale data measured in street canyons at four European cities. Thesensitivity of SEUS model is studied analytically. Results show that relative errors in the evaluation of thetwo dimensionless empirical parameters have less inuence on model uncertainties than uncertainties1874 Buenos Aires, Argentina

National Scientic and Technological Research Council (CONICET), Avellaneda Regional Faculty, National Technological University, Av. Ramn Franco 5050,A simple model for calculating air pollu

Laura E. Venegas*, Nicols A. Mazzeo, Mariana C. DAll rights reserved.n within street canyons

zutti

le at ScienceDirect

nvironment

vier .com/locate/atmosenv

(m s1) is the trafc velocity, a and b are dimensionless empiricalparameters. Parameter a depends, among other factors, on streetgeometry, wind direction and sampling position. The height of thebuildings is implicitly included in Eq. (1) through parameter a aspart of the street geometry. Parameter b depends on trafc owconditions. Turbulent motions mechanically generated by trafcbecome an important factor for the dilution of pollutants in streetsunder lowwind speed conditions. Empirical expressions for a and bhave been derived using full-scale experimental data from fourstreet canyons (Mazzeo and Venegas, 2010, 2011; Mazzeo et al.,2012). Available information used to obtain functional forms for aand b includes hourly air pollutant concentrations and trafc ow

ic Environment 87 (2014) 77e86Several dispersion models have been specially developed for orsimply used in street canyons applications. They can be useful in airquality and trafc management, urban planning, interpretation ofmonitoring data, pollution forecasting, population exposure, etc.(e.g. Ganguly and Broderick, 2010; Holmes and Morawska, 2006;Jensen et al., 2009; Kukkonen et al., 2003; Vardoulakis et al.,2003; Wang and Xie, 2009). Air quality models can be dividedinto two main categories: parametric and numerical models.Among the rst group, mathematical techniques can be simpleparameterisations such as the STREET (Johnson et al., 1973) and theSTREET-BOX (Mensink and Lewyckyj, 2001) models or moredeveloped models as the Canyon Plume-Box Model (CPBM)(Yamartino and Wiegand, 1986), the Assessing the Environment OfLocations In Urban Streets (AEOLIUS) model (Buckland andMiddleton, 1999), the Danish Operational Street Pollution Model(OSPM) (Berkowicz, 2000; Berkowicz et al., 2008), the Semi-Empirical Parameterized Street Canyon Model (SEP-SCAM)(Papathanassiou et al., 2008) and the parametric relations in thestreet network approach (Soulhac et al., 2013). In the second group,numerical approaches include computational uid dynamics (CFD)(Eichhorn,1996; Murena et al., 2009; Parra et al., 2010; Schatzmannand Leitl, 2011; Solazzo et al., 2011; Stern and Yamartino, 2001) andlarge-eddy simulations (LES) (Li et al., 2012; Moonen et al., 2013;Zhang et al., 2011). Simple models require smaller amount ofinput information and computational resources than moreadvanced simulation techniques. Nevertheless, they do not takeinto account certain characteristics, such as dispersion processesrelated to the conguration of buildings in urban intersections, theshape of building roofs, etc. On the other hand, numerical modelsproduce detailed modelling outputs in terms of ow and turbu-lence and can be used to study the air ow structure inside thestreet canyon.

This paper introduces the Semi-Empirical Urban Street (SEUS)model which includes new empirical parameterisations of windrelated and trafc-induced turbulences in street canyons. SEUS is asimple parameterised model to estimate inert air pollutants andnitrogen dioxide (NO2) concentrations using chemical processes instreet canyons. The sensitivity of SEUS estimations is studied usingthe propagation of errors methodology. SEUS is applied to computenitrogen oxides (NOx) and NO2 hourly concentrations in a streetcanyon in the city of Buenos Aires using historical trafc informa-tion, routine meteorological data and modelled urban backgroundconcentrations. Statistical evaluation of model results is donecomparing estimated hourly concentrations with observationaldata and with results of a widely used operational street canyonmodel.

2. The SEUS model

2.1. Model description

The Semi-Empirical Urban Street (SEUS) model is based on thescaling of air pollution concentration (C) inside urban street can-yons given by the following expression:

C E u1s W1 Cb (1)

where E is the emission rate per length, W is the width of thecanyon, Cb is the background concentration and us is the dispersivevelocity scale given by Kastner-Klein et al. (2003):

us s2u s2v

1=2 aU2 bV21=2 (2)where s2 aU2 is the wind speed variance, s2 bV2 is the trafc-

L.E. Venegas et al. / Atmospher78u v

induced velocity variance, U (m s1) is the ambient wind speed, Vdata registered at the following street canyons (www2.dmu.dk/AtmosphericEnvironment/trapos/datadoc.htm): Gttinger Strasse(Hannover, Germany), Schildhornstrasse (Berlin, Germany), Jagtvej(Copenhagen, Denmark) and Hornsgatan (Stockholm, Sweden)along with wind speed and direction data registered nearby. Oneyear of hourly information was available for each street canyon. Asthe street canyons have different orientations, it is necessary tointroduce a new parameter (q) to refer the results of the four streetcanyons to a common value relative to the street orientation, thewind direction (WD) and the position of the monitoring stationwithin the street. Parameter q (Fig. 1) is dened as

q WD ST for WD STWD 360 ST for WD < ST (3)

where ST is the angle between the north and the street axis towardsthe right side of the monitoring location (facing the street). Thevalue of q is expressed in degrees. In this way, if wind blowsperpendicular to the street axis, the leeward situation veriesq 90 and the windward situation q 270, in any street canyon.

In the evaluation of parameters a and b, full-scale data havebeen grouped into 16 wind sectors of 22.5 centred in q 0, 22.5,45.0, 67.5, 90.0, 112.5, 135.0, 157.5, 180.0, 202.5, 225.0,247.5, 270.0, 292.5, 315.0 and 337.5. The values of a and bobtained for each street canyon applying statistical methods toexperimental data are presented in Mazzeo and Venegas (2005,2010, 2011).

The variation of a (averaged over the four street canyons) with qis presented in Fig. 2. The tting curve (coefcient of determination,R2 0.951) is given by

a 0:002745 exp0:452317 1:9803 sin1:01821pq=180(4)

valid for 0 q < 360.The variations of b with trafc density (nv) and q for the four

street canyons have been studied in Mazzeo and Venegas (2011).The trafc density is nv (km1) Nt/V where Nt (h1) is the totalnumber of vehicles passing the street per time unit given byFig. 1. Denition of parameter q.

The sensitivity of estimated concentrations to variations in themagnitude of each parameter included in Eq. (6) is studied using

L.E. Venegas et al. / Atmospheric ENt Np Nh, Np (h1) is the number of small vehicles per hour(passenger cars and light duty vehicles), Nh (h1) is the number oflarge vehicles per hour (heavy duty vehicles, buses and trucks) andV is calculated as V (km h1) (Np Vp Nh Vh)/(Np Nh) where Vp(km h1) is the velocity of small vehicles and Vh (km h1) is that oflarge vehicles.

Considering the cases with 0 q 180, b increases with trafcdensity but it does not show a clear dependence on q. In order toobtain the empirical expression of b as a function of nv, the values ofb have to be sorted on nv into the following bins: 120 km1 because the number ofcases of b is not uniformly distributed with trafc density (Mazzeoand Venegas, 2011; Mazzeo et al., 2012). Fig. 3 shows the variationof b (averaged for each bin) with trafc density and the best ttingcurve (R2 0.925) given by

b 2:88642E 06 nv 0:930771 (5)

valid for 0 q 180, the unit of the numerical coefcient2.88642E-06 (km0.930771) results from the tting methodology andit is such that dimensionless value of b is obtained for nv expressedin km1 (Mazzeo et al., 2012). Trafc induced turbulence (sv2 bV2)is estimated using Eq. (5) and V taking into account trafccomposition, but the different contributions generated by vehiclesof different sizes are not explicitly identied. Mazzeo et al. (2012)used this expression and three other parameterisations of trafc-produced turbulence proposed by other authors into the scaling

Fig. 2. Variation of a with q.of air pollutants inside a street canyon to estimate hourly nitrogenoxides (NOx) concentrations at ground-level for leeward conditions

Fig. 3. Variation of b with trafc density.the propagation of errors methodology. Because of the simple formof Eq. (6), it is possible to examine its sensitivity analytically, bymeans of the partial derivate of the calculated concentration (C)with respect to each parameter p (Hilst, 1970). In this way, thesensitivity of model estimations jdC/Cj to variations in the magni-tude of each parameter (p E, W, U, V, a, b, Cb) is given by jdC/Cj Kjdp/pj. The form of theweighting factor (K) for each parametercan be obtained from:

dCC

dEE

dWW

aU2

aU2 bV2$dUU

bV

2

aU2 bV2$dVV

aU2

2aU2 bV2

$daa

bV2

2aU2 bV2

$dbb

dCbCb

(7)

As expected, relative variations in E, W and Cb produce similarand compared results with observations. The results obtained usingb given by Eq. (5) were as good as obtained considering otherparameterisations (which account for the differences in turbulencegenerated by vehicles of different size).

If 180< q < 360 and U > 2 m s1, the receptor receives the re-circulating part of the pollutants in the street and in these cases thecontribution of trafc-induced turbulence to the dilution of pol-lutants can be neglected, so b 0. However, for meteorologicalconditions with 180< q < 360 and lowwind speeds (U 2 m s1)b is evaluated using Eq. (5).

Eqs. (1)e(5) constitute SEUS model for inert pollutants. Insummary, the application of SEUSmodel to estimate inert pollutantconcentrations requires the following input information: streetcanyon dimensions, trafc ow and composition in the street, airpollutant emissions, ambient wind speed and direction, and back-ground concentration contribution from sources outside the street.

2.2. Sensitivity analysis of the basic expression of SEUS

Combining Eqs. (1) and (2), the basic expression of SEUS is givenby

C EaU2 bV2

1=2W1 Cb (6)

Table 1Range of the weighting factor K [Eq. (7)] of the parameters included in Eq. (6), fordifferent intervals of q.

q [0e45] (45e90] (90e135] (135e180] (180e360)

U 0.006e0.99 0.003e0.99 0.004e0.99 0.006e0.99 1.0V 0.001e0.99 0.004e0.99 0.003e0.99 0.001e0.99 ea 0.003e0.50 0.0015e0.50 0.002e0.50 0.003e0.50 0.50b 0.0005e0.50 0.002e0.50 0.0015e0.50 0.0005e0.50 e

nvironment 87 (2014) 77e86 79relative variations in C as K 1.0 for these variables. For the otherparameters (U, V, a, b) K varies with q. The ranges of the weightingfactor of the relative variation of these variables for different in-tervals of q are included in Table 1. For 180< q< 360, K 1 for dU/U and K 0.5 for da/a. For 0 q 180, relative variations dU/U anddV/V produce less equal relative variations in C because0.001 K 0.99. Eq. (7) shows that the relative variations in a andb have smaller inuence on dC/C than the variation of other vari-ables because da/a and db/b are multiplied by K 0.5 [aU2/(aU2 bV2)] and K 0.5 [bV2/(aU2 bV2)], respectively. The eval-uation of these weighting factors (Table 1) shows that relative

orientated eastewest is summarised in Fig. 4. The model is runconsidering that the street canyon is 20 m wide with buildings of20 m height and the receptor is located on the southern side of thestreet. Computations are made for a constant trafc ow consistingof 900 light and 100 heavy vehicles per hour, travelling at a speed of40 km h1. The emission rate is set to 1000 ppb m2 s1. The back-ground concentration is Cb 0 ppb and chemical reactions are nottaken into account.

Fig. 4 shows that SEUS clearly reproduces higher concentrationson leeward conditions than on windward conditions. The differ-ence between them decreases for lower wind speeds. In calms, theturbulence created by trafc, determines the dispersion conditions.

This example serves as an illustration of the model behaviouronly.

3. Results and discussion

ic Environment 87 (2014) 77e86variations in a and b have the smallest inuence on dC/C as for theseparameters 0.0005 K 0.50.

2.3. Chemical processes in the street canyon

Due to the relatively short time scales involved, only fastchemical reactions can have a signicant inuence on air compo-sition in a street canyon. For nitrogen oxides, it is therefore suf-cient to include only the basic reactions involving nitrogenmonoxide (NO), nitrogen dioxide (NO2), oxygen (O2) and ozone(O3). SEUS includes the same simplied chemistry as in the OSPMmodel (Berkowicz et al., 1997). For practical purposes, the threebasic reactions between the mentioned species are:

NO O3/NO2 O2 (8a)

NO2 hy/NO O, (8b)

O, O2 M/O3 M (8c)

whereM (usually N2 or O2) represents a molecule that absorbs theexcess vibrational energy and thereby stabilises the O3 moleculeformed, hy represents the energy of a photon and O is the activemono-atomic oxygen. The reaction between the oxygen radical (O

and the molecular oxygen (O2) is so fast that for all practical pur-poses the reaction system can be restricted to reactions (8a) and(8b). Taking into account the chemical transformations and theexchange of street canyon air with the ambient air, under steady-state conditions, the total concentration of NO2 in the street iscalculated by Berkowicz et al. (1997):

NO2 0:5B

B2 4C

1=2(9)

with

B NOx J=k 1=ks NO2v NO2b O3b

C NOxNO2v NO2b O3b NO2v NO2bks

[NOx] is the total concentration of NOx in the street; [NO2]b and[O3]b refer to background concentrations of NO2 and O3, respec-tively; [NO2]v ( f [NOx]v) is the contribution from the directemission in the street, f is the fraction of NOx directly emitted asNO2 and [NOx]v is calculated by the rst term in Eq. (6)([NOx] [NOx]v [NOx]b); k (ppb1 s1) 5.38E-02 exp(1430/T1) is the reaction coefcient of Eq. (8a) which is a function of airtemperature, T (K); J is the photo-dissociation coefcient that de-pends on global radiation, q (Wm2), and is given by J (s1) 8.0E-04 exp(10/q) (7.4E-06)q if q 1.0Wm2, otherwise J 0; s H/swt, refers to pollutants residence time in the street and is calcu-lated as the ratio between the depth of the street canyon (given bythe average building height, H) and the turbulent exchange velocityswt [0.01U2 0.4sv2]1/2.

In case of modelling NO2 concentrations, the fraction of NOxdirectly emitted as NO2, air temperature and global radiation, andbackground concentrations of NOx, NO2 and O3 are required inaddition to the input information already mentioned for inertspecies.

2.4. Study of the behaviour of SEUS with ambient wind

The variation of air pollutant concentrations with ambient wind

L.E. Venegas et al. / Atmospher80conditions, estimated by SEUS in a hypothetical street canyon3.1. Site and data

SEUS model has been applied to estimate NOx and NO2 hourlyconcentrations in a street canyon of Crdoba Avenue (BuenosAires). This street canyon is busy, irregular and asymmetric (Fig. 5).Crdoba Ave. has ve lanes, is 30 m wide and is orientated east towest. Buildings on the northern side of the avenue are low (w10m)and quite regular. On the southern side, buildings are much tallerand of different heights (10 me80 m). Trafc ow in the avenue isapproximately 38000 veh day1 with a mean fraction of large ve-hicles (buses and trucks) of 5%. Average trafc speed is about30 km h1. Fig. 6 illustrates the diurnal variation of hourly averagetrafc ow for working days. Information on trafc ow, averageroad trafc speed and vehicle eet composition has been obtainedfrom several reports elaborated by local authorities (GCBA, 2006;GCBA-ACOM, 2006). Unfortunately there is no actual informationon trafc ow for the period of study. Model estimations arecompared with measurements obtained from a monitoring stationat the southern side of the canyon near 10 m east from the inter-section with R. Pea St. Trafc ow in Crdoba Ave. is more thanthree times greater than the trafc ow in the crossing street and itcan be expected to be the main responsible for observed airpollutant concentrations at the monitoring site. The monitoringstation is located at a kerbside site in a commercial area withbuildings destined to business activities and residence. It integratesthe air quality monitoring network of the city and belongs to theEnvironmental Protection Agency of the city of Buenos Aires. At thissite, continuous air quality measurements cover CO, NO, NOx, NO2,Fig. 4. Variation of estimated concentration with ambient wind direction and speed(U) at a south side receptor in an east-west oriented street canyon.

ation (C) in Crdoba Ave., Buenos Aires. (Picture: Google Earth).

L.E. Venegas et al. / Atmospheric Environment 87 (2014) 77e86 81SO2 and PM10. Themeasuring height is 2.5m and the distance fromthe building walls is 5 m. The concentration of nitrogen oxides wasanalysed by a chemiluminescence monitor. The accuracy of themonitor was better than 5% at a concentration of 200 ppb for NO.The precision was 1% and the linearity was /1% full scale.

SEUS is run using the hourly meteorological information regis-tered at the meteorological station of the National MeteorologicalService in the domestic airport located in the city, near the coastabout 3.5 km NNE of the monitoring site. Due to lack of availableinformation on ambient wind at the monitoring site wind speedand direction registered at the domestic airport have beenconsidered as ambient wind conditions (Mazzeo and Venegas,2012). It is recommended to include observed background con-centrations in calculations. However, in the city of Buenos Airesthere is not a continuous air quality monitoring station represen-tative of urban background concentrations. Several experimentalcampaigns (Bogo et al., 1999, 2001; Mazzeo et al., 2005) reportedlow levels of ozone concentration in Buenos Aires. Model calcula-tions are done assuming a value of 23ppb for background ozoneconcentration based on the background ozone concentration ob-tained during an experimental campaign and discussed in Mazzeo

Fig. 5. Picture of a 3D perspective of the monitoring site locet al. (2005). On the other hand, background concentrations of NOxand NO2 considered in calculations have hourly variations.Modelled hourly background concentrations of NOx and NO2 areestimated applying the DAUMOD urban atmospheric dispersionmodel (Pineda Rojas and Venegas, 2013; Venegas and Mazzeo,2006) in the Metropolitan Area of Buenos Aires considering anhourly emission inventory with a grid resolution of 1 km2 (Venegaset al., 2011). In order to avoid double-counting of emissions at the

Fig. 6. Diurnal variation of hourly average trafc ow for working days.

Fig. 7. Comparison between estimated and observed NOx (a) and NO2 (b) hourlyconcentrations (ppb) categorised into two wind speed (U) classes. Lines showing anagreement between estimations and data by a factor of two are also indicated.

comparisons of estimated and observed hourly concentrations ofNOx and NO2 are presented in Fig. 7. The data have been categorisedinto two ambient wind speed (U) classes, U > 2 m s1 andU 2 m s1. The scatter of observed and estimated data points iswider for low wind speed (U 2 m s1) conditions compared withthe cases with higher wind speeds (U > 2 m s1), for both NOx andNO2. The internal variation of data points in both scatter plots issubstantial. However, for both pollutants, a great portion of datapoints is located inside the factor of two lines.

According to the scatter plots (Fig. 7) the model underestimatesthe highest concentration values of both pollutants. However, it isalso of interest to nd out whether the model can generate aconcentration distribution that is similar to the observed, especiallyat the upper percentiles. Fig. 8aeb show comparisons of estimatedand observed hourly concentration distributions for NOx and NO2,respectively. Notice that in these plots estimated and observedconcentrations for the same percentile may not correspond to thesame case nor occur under the same conditions. For both pollut-ants, Fig. 8 illustrates the overestimation of model results at lowconcentrations, a good agreement at high percentiles and the un-derestimation of extremely high values. This is of particular interest

L.E. Venegas et al. / Atmospheric Environment 87 (2014) 77e8682Fig. 8. Estimated and observed cumulative frequency distributions of NOx (a) and NO2(b) hourly concentrations.receptor, hourly variations of NOx and NO2 background concen-trations at the receptor grid are obtained averaging the values atthe surrounding grids. The direct emission of NO2 is considered 10%of NOx (Pineda Rojas and Venegas, 2013).

3.2. Comparison of model results with observations

Seven month of measured hourly concentrations of nitrogenoxides (NOx) and nitrogen dioxide (NO2) are available. The

Table 2Summary of estimated and observed hourly concentrations for NOx and NO2 and their agcoefcient; FA2: fraction of modelled values between a factor 2 of observations; FB: frac

All data U > 2 m s1 U N 4488 3893 595

Results for NOxMeanobs (ppb) 93.7 88.7 126.4Meanest (ppb) 103.8 99.1 134.8Sigmaobs (ppb) 70.2 66.1 85.7Sigmaest (ppb) 61.1 56.4 78.6NMSE 0.34 0.31 0.4R 0.631 0.659 0.4FA2 0.727 0.735 0.6FB 0.103 0.111 0.0Results for NO2Meanobs (ppb) 32.8 31.9 38.7Meanest (ppb) 35.5 34.0 44.9Sigmaobs (ppb) 13.2 12.8 14.2Sigmaest (ppb) 11.6 10.6 12.9NMSE 0.14 0.13 0.1R 0.512 0.506 0.4FA2 0.913 0.919 0.8FB 0.078 0.064 0.1for hourly NO2 concentrations because in several countries airquality regulations for NO2 usually involve the value of the pthpercentile (for example, the 98th percentile in the NationalAmbient Air Quality Standards for the US and in the EnvironmentalLaw for the city of Buenos Aires in Argentina, or the 18 permittedexceedences each year-equivalent to the 99.8th percentile- in theEU).

The overestimation of low values suggests an overestimation ofbackground concentrations (Cb). Estimated NOx and NO2 hourlyconcentrations at the lower 10th-percentile range in Fig. 8aeb aresensitive to Cb. In this range, Cb contributed with more than 50% ofthe estimated concentration in 96% of the NOx and in 98% of theNO2 hourly values.

The evaluation of the agreement between SEUS estimations andobservations is done considering all data and also regarding: a)wind speed: U > 2 m s1 and U 2 m s1 (including calms), and b)wind direction: leeward (22.5< q 157.5), windward(202.5< q 337.5) and parallel (337.5 < q 360.0;0.0 < q 22.5; 157.5 < q 202.5) situations. Table 2 summa-rises the statistical evaluation of model performance including thenumber of values (N), the mean and sigma of estimated (Ce) and

reement. (N: number of cases; NMSE: normalised mean square error; R: correlationtional bias).

2 m s1 Leeward Windward Parallel

1229 1725 1390

130.7 64.2 95.3122.1 81.8 112.076.1 53.9 64.469.0 47.3 57.6

3 0.18 0.54 0.3467 0.732 0.509 0.54874 0.861 0.630 0.74265 0.067 0.241 0.161

36.1 29.5 33.537.9 31.8 36.612.8 13.3 12.512.3 10.3 10.7

5 0.11 0.16 0.1309 0.542 0.488 0.45874 0.945 0.881 0.934

48 0.048 0.075 0.087

clearly reproduced by SEUS, although the observed leeward/windward concentration ratio is higher than the modelled ratio forthe perpendicular wind condition. Discrepancies between esti-mated and observed mean concentration values are more pro-nounced for wind directions from the sector N e ESE. Buildingheights in the street canyons of the European cities, on which themodel development has been based, are more regular than inBuenos Aires. The existence of irregular building heights andasymmetries in the street canyon where the model is applied mayenhance discrepancies between estimations and observations.

The pollution roses for NO2 show a little variation of meanconcentrations with wind direction. This variation can be explainedby the fact that NO2 is mainly a secondary pollutant. In general, thepatterns of estimated and observed pollution roses are quitesimilar. Discrepancies between both pollution roses are more pro-

Fig. 9. Comparison between observed and estimated air pollution roses for NOx (a) andNO2 (b). The heavy straight line indicates the street canyon orientation. The monitoringsite is on the southern side of the street.

ric Eare 0.732 and 0.542, for NOx and NO2 estimations, respectively. Forboth pollutants, the values of NMSE and FB are low.

On the other hand, the model overestimates the hourly con-centrations of both pollutants under windward conditions, with FBvalues of 24.1% for NOx and 7.5% for NO2.

The observed and estimated pollution roses for NOx and NO2 areshown in Fig. 9aeb, respectively. The pollution roses for NOx show aclear dependence of concentrations on wind direction. The streetThe analysis of SEUS results regarding thewind direction revealsthat the best performance of the model is obtained for leewardcases. In these conditions, 86.1% of NOx and 94.5% of NO2 estima-tions are between a factor of two the observations. The FB valuesare 6.7% and 4.8%, the NMSE are 18% and 11% and the R valuesvalues are 6.4% and 14.8% for the scatter plots corresponding tothe higher wind speed conditions and lower wind speed condi-tions, respectively. For lower wind speed cases, model estimationsof both pollutants show larger relative errors and weaker correla-tions compared with the results for higher wind speed.wind speeds. For NOx, the NMSE values are 43% and 31%, the Rvalues are 0.467 and 0.659 and the FB values 6.5% and 11.1% forthe estimations obtained for low wind speed (U 2 m s1) and forU > 2 m s1, respectively. On the other hand, for NO2 the NMSEvalues are 13% and 15%, the R values are 0.506 and 0.409 and the FBobserved (Co) values, and the following measures (Chang andHanna, 2004):

- Normalised Mean Square Error

NMSE Co Ce2.

Co Ce

- Correlation coefcient

R Co Co

Ce Ce

.sCe sCo

- Fraction of estimations within a factor of two of the observations(FA2)

0:5 Ce=Co 2:0

- Fractional bias

FB 2Co Ce

.Co Ce

A perfect model would have R 1.0; FA2 1.0; NMSE 0.0 andFB 0.0. Chang and Hanna (2004) based on extensive experiencewith evaluating many models with many eld data sets, summar-ised typical magnitudes of the above performance measures ofmodel acceptance criteria. They concluded that acceptable per-forming models have the following typical performance measures:FA2 > 0.5, 0.3 < FB < 0.3 and NMSE < 1.5. However, these are notrm guidelines and it is necessary to consider all performancemeasures in making a decision concerning model acceptance.

Considering all data, 72.7% of NOx and 91.3% of NO2 estimatedvalues are between a factor of two of observations. The FB values forthe scatter plots of NOx and NO2 are10.3% and7.8%, respectively.The NMSE values are 34% and 14% for NOx and NO2 estimations,respectively. These results showan acceptable performance of SEUSin estimating hourly concentrations of both pollutants in thisirregular street canyon, when using routine meteorological infor-mation and modelled background concentrations.

Regarding wind speed, the statistics included in Table 2 show abetter model performance for U > 2 m s1 than for lower ambient

L.E. Venegas et al. / Atmosphecanyon effect, i.e. higher leeward than windward concentrations, isnvironment 87 (2014) 77e86 83nounced for wind direction WNW, NE and ENE.

Additionally, trafc emissions in the crossing street may impactparticularly on NOx concentrations and contribute to the higherobserved mean concentration for westerly wind direction shown inFig. 9a. In this direction, the estimated averaged NOx concentrationis 14% less than that observed. The impact of the emissions in thecrossing street on NO2 concentrations registered at the monitoringsite is expected to be small due to the secondary nature of this airpollutant (as can be seen in Fig. 9b). The better match betweenobserved and estimated pollution roses for NO2 than for NOx isreected in the lower fractional bias (FB) for NO2 than for NOx(Table 2).

The comparison of daily variations of hourly mean estimatedand observed concentrations for NOx and NO2 are presented inFig. 10aeb. Diurnal variations are plotted for Mondays to Fridays,Saturdays and Sundays. In general, the shape of diurnal and weeklyvariations is well reproduced. Uncertainties in trafc ow inputdata may explain the departures between estimated and observedmean values. Mean hourly concentrations are overestimated atrush-hour in the evening and underestimated during the night.However, from the averaged values shown in Fig. 10 it is notpossible to infer any conclusion on model performance in esti-mating hourly concentrations. This is because hourly concentra-tions observed at the same hour during the period of study, show awider spread than hourly estimations for that hour. Therefore, theaverage of hourly estimations can be greater than the average ofhourly observations (as at rush-hour in the evening) whereas for

OSPM is applied to estimate NOx and NO2 hourly concentrations inCrdoba Ave. for the same period using the same input information(except for the detailed street conguration) as in SEUS application.The statistical measures to evaluate the performance of OSPM forall data are summarised in Table 3. These values and the obtainedfor the proposed model (rst column of Table 2) show goodagreement between the estimations of both models and observa-tions. Also, their performances are quite similar with NMSE of 0.34(SEUS) and 0.42 (OSPM) for NOx, and 0.14 (SEUS) and 0.16 (OSPM)for NO2.

Moreover, Fig. 11aeb show the excellent agreement betweenSEUS and OSPM estimated NOx and NO2 hourly concentrations atstreet level in Crdoba Ave. The statistical evaluation of SEUS es-timations when compared to OSPM results is summarised in

Table 3Summary of OSPM estimated and observed NOx and NO2 hourly concentrations andtheir agreement (N: number of cases; NMSE: normalised mean square error; R:correlation coefcient; FA2: fraction of modelled values between a factor 2 of ob-servations; FB: fractional bias).

N 4488 Mean (ppb) Sigma (ppb) NMSE R FA2 FBNOx_obs 93.7 70.2 e e e eNOx_OSPM 116.5 65.6 0.42 0.565 0.677 0.217NO2_obs 32.8 13.2 e e e eNO2_OSPM 38.4 12.2 0.16 0.486 0.886 0.157

L.E. Venegas et al. / Atmospheric Environment 87 (2014) 77e86843.3. Comparison between SEUS and OSPM performances in CrdobaAve

The performance of SEUS in Crdoba Ave. is compared with thatof the well-known widely used Windows-based version of OSPM.those hours the highest estimated hourly concentration is lowerthan the observed one (upper percentiles in Fig. 8). On the otherhand, at night the model underestimates the averaged values(Fig. 10) but the lowest hourly estimation is greater than the lowestobserved value (lower percentiles in Fig. 8).Fig. 10. Average diurnal variation of estimated and observed NOx (Table 4.With NMSE of 0.04 for NOx and 0.01 for NO2, there has beenno signicant discrepancy between the results of both modelswhen using the same input information.

4. Conclusions

This paper describes the simple mathematical street canyonmodel SEUS which includes empirical expressions for the param-eterisation of turbulent motions related towind and trafc recentlyobtained from full-scale data. The parameter included in theparameterisation of wind-induced dispersive motions is expressedas a function of wind direction, street orientation and samplinga) and NO2 (b) concentrations in Crdoba Ave. (Buenos Aires).

L.E. Venegas et al. / Atmospheric Elocation within the street. The parameter included in the parame-terisation of trafc-induced turbulence is expressed in function oftrafc density (trafc ow and composition and trafc velocity).The sensitivity analysis of SEUS model indicates that the un-certainties in the evaluation of the two empirical parameters haveless inuence on model uncertainties than that of the uncertaintiesin the other input variables.

toring data are not available, it is recommended that more than one

the city of Buenos Aires and the National Meteorological Service of

Fig. 11. Comparison between SEUS and OSPM estimated NOx (a) and NO2 (b) hourlyconcentrations in Crdoba Ave. (Buenos Aires).

Table 4Summary of SEUS and OSPM estimations of NOx and NO2 hourly concentrations andtheir agreement. (N: number of cases; NMSE: normalised mean square error; R:correlation coefcient; FA2: fraction of modelled values between a factor 2 of ob-servations; FB: fractional bias).

N 4488 Mean (ppb) Sigma (ppb) NMSE R FA2 FBNOx_OSPM 116.5 65.6 e e e eNOx_SEUS 103.8 61.1 0.04 0.958 1.00 0.115NO2_OSPM 38.4 12.2 e e e eNO2_SEUS 35.5 11.6 0.01 0.972 1.00 0.079Argentina for data used in this study.

References

Berkowicz, R., Hertel, O., Larsen, S.E., Sorensen, N.N., Nielsen, M., 1997. ModellingTrafc Pollution in Streets. Ministry of Environment and Energy, NERI.

Berkowicz, R., 2000. OSPM e a parameterised street pollution model. Environ.Monit. Assess. 65, 323e331.

Berkowicz, R., Ketzel, M., Jensen, S., Hvidberg, M., Raaschou-Nielsen, O., 2008.Evaluation and application of OSPM for trafc pollution assessment for a largenumber of street locations. Environ. Model. Softw. 23, 296e303.

Bogo, H., Negri, R.M., San Romn, E., 1999. Continuous measurement of gaseouspollutants in Buenos Aires city. Atmos. Environ. 33, 2587e2598.

Bogo, H., Gmez, D.R., Reich, S.L., Negri, R.M., San Romn, E., 2001. Trafc pollutionin a downtown site of Buenos Aires city. Atmos. Environ. 35, 1717e1727.

Britter, R.E., Hanna, S.R., 2003. Flow and dispersion in urban areas. Annual Rev. FluidMech. 35, 469e496.

Buckland, A.T., Middleton, D.R., 1999. Nomograms for calculating pollution withinstreet canyons. Atmos. Environ. 33, 1017e1036.

Chang, J.C., Hanna, S.R., 2004. Air quality model performance evaluation. Meteorol.Atmos. Phys. 87, 167e196.

Eichhorn, J., 1996. Validation of a microscale pollution dispersal model. In: AirPollution Modeling and its Application IX. Plenum Press, New York.

Ganguly, R., Broderick, B.M., 2010. Estimation of CO concentrations for an urbanstreet canyon in Ireland. Air Qual. Atmos. Health 3, 195e202.

GCBA, 2006. Gobierno de la Ciudad de Buenos Aires. In: Informes sobre ndice deTrnsito (in Spanish).

GCBA-ACOM, 2006. Mapa Estratgico de Ruido-Ciudad Autnoma de Buenos Aires(Strategic Noise Map-City of Buenos Aires). Gob. de la Ciudad Autnoma deBuenos Aires y Asoc. Civil Oir Mejor (in Spanish).

Hilst, G.R., 1970. Sensitivities of air quality prediction to input errors and un-certainties. In: Proc. Of Symposium on Multiple-source Urban Diffusion Models.US. EPA, Research Triangle Park, NC.

Holmes, N.S., Morawska, L., 2006. A review of dispersion modelling and its appli-cation to the dispersion of particles: an overview of different dispersion modelsavailable. Atmos. Environ. 40, 5902e5928.model should be used to avoid excessive reliance on a specic code.

Acknowledgements

This study has been supported by Project CONICET-PIP 0304.The authors wish to thank the Environmental Protection Agency ofSEUS can be applied to estimate ground-level concentrationsinside street canyons for any non-reacting pollutant if data con-cerning trafc emissions and urban background concentrations areknown. It can also estimate NO2 concentrations as it takes intoaccount a simple photochemistry between NO, NO2 and O3.

SEUS has been applied to estimate NOx and NO2 hourly con-centrations in an irregular street canyon in the city of Buenos Airesusing historical trafc information, routinemeteorological data andmodelled urban background concentrations. The results have beencompared with observations obtained from the station of theEnvironmental Protection Agency of the city of Buenos Aires in thiscanyon. For both pollutants, the statistical evaluation shows a goodagreement between estimated and observed hourly concentra-tions. SEUS model has been reasonably accurate, in spite of itssimplicity. Furthermore, the performances of SEUS and OSPMmodels in Crdoba Ave. using the same input data are very similar.

SEUS requires a relative small amount of input information anddue to its simplicity it can be easily implemented on a standardspreadsheet which can be used to obtain a rst estimation of the airquality at a ground-level receptor inside a street canyon. It couldbecome a useful screening tool in air quality applications, if suc-cessful evaluations in a wide variety of urban street canyons areaccomplished. Future research will aim at validations of the pro-posed model for more air quality data obtained in different streetcanyons.

In air quality assessment at street level, when relevant moni-

nvironment 87 (2014) 77e86 85Jensen, S.S., Larson, T., Deepti, K.C., Kaufman, J., 2009. Modeling trafc air pollutionin street canyons in New York City for intra-urban exposure assessment in the

US multi-ethnic study of atherosclerosis and air pollution. Atmos. Environ. 43,4544e4556.

Johnson, W.B., Ludwig, F.L., Dabberdt, W.F., Allen, R.J., 1973. An urban diffusionsimulation model for carbon monoxide. J. Air Pollut. Control Assoc. 23, 490e498.

Kastner-Klein, P., Fedorovich, E., Ketzel, M., Berkowicz, R., Britter, R.E., 2003. Themodelling of turbulence from trafc in urban dispersion models e part II.Evaluation against laboratory and full-scale concentration measurements instreet canyons. Environ. Fluid Mech. 3, 145e172.

Kukkonen, J., Partanen, L., Karppinen, A., Walden, J., Kartastenp, R., Aarnio, P.,Koskentalo, T., Berkowicz, R., 2003. Evaluation of the OSPM model combinedwith an urban background model against the data measured in 1997 in Run-enberg Street, Helsinki. Atmos. Environ. 37, 1101e1112.

Li, X.-X., Britter, R.E., Norford, L.K., Koh, T.-Y., Entekhabi, D., 2012. Flow and pollutanttransport in urban street canyons of different aspect ratios with ground heat-ing: large-Eddy simulation. Bound. Layer Meteorol. 142, 289e304.

Mazzeo, N.A., Venegas, L.E., 2005. Evaluation of turbulence from trafc usingexperimental data obtained in a street canyon. Int. J. Environ. Pollut. 25, 164e176.

Mazzeo, N.A., Venegas, L.E., 2010. Air pollution dispersion inside a street canyon ofGttinger Strasse (Hannover, Germany): new results of the analysis of full scaledata. Int. J. Environ. Pollut. 40, 195e209.

Mazzeo, N.A., Venegas, L.E., 2011. Study of natural and trafc-producing turbulencesanalysing full-scale data from four street canyons. Int. J. Environ. Pollut. 47,290e301.

Mazzeo, N.A., Venegas, L.E., 2012. Hourly NOx concentrations and wind direction inthe vicinity of a street intersection. Int. J. Environ. Pollut. 48, 96e104.

Mazzeo, N.A., Venegas, L.E., Choren, H., 2005. Analysis of NO, NO2, O3 and NOxconcentrations measured at a green area of Buenos Aires city during winter-time. Atmos. Environ. 39, 3055e3068.

Mazzeo, N.A., Venegas, L.E., Dezzutti, M.C., 2012. Air pollution in a street canyonestimated considering different parameterisations of vehicle-induced turbu-lence. Int. J. Environ. Pollut. 50, 120e129.

Mensink, C., Lewyckyj, N., 2001. A simple model for the assessment of air quality instreets. Int. J. Veh. Des. 27, 242e250.

Moonen, P., Gromke, C., Dorer, V., 2013. Performance assessment of Large EddySimulation (LES) for modelling dispersion in an urban street canyon with treeplanting. Atmos. Environ. 75, 66e76.

Murena, F., Favale, G., Vardoulakis, S., Solazzo, E., 2009. Modelling dispersion oftrafc pollution in a deep street canyon: application of CFD and operationalmodels. Atmos. Environ. 43, 2303e2311.

Oke, T.R., 1996. Boundary Layer Climates, second ed. Routledge, London.Parra, M.A., Santiago, J.L., Martn, F., Martilli, A., Santamara, J.M., 2010.

A methodology to urban air quality assessment during large time periods ofwinter using computational uid dynamic models. Atmos. Environ. 44, 2089e2097.

Papathanassiou, A., Douros, I., Moussiopoulos, N., 2008. A simplied three-dimensional approach to street canyon modelling using SEP-SCAM. Environ.Model. Softw. 23, 304e313.

Pineda Rojas, A., Venegas, L., 2013. Upgrade of the DAUMOD atmospheric dispersionmodel to estimate urban background NO2 concentrations. Atmos. Res. 120-121,147e154.

Schatzmann, M., Leitl, B., 2011. Issues with validation of urban ow and dispersionCFD models. J. Wind Eng. Ind. Aerodyn. 99 (4), 169e186.

Solazzo, E., Vardoulakis, S., Cai, X., 2011. A novel methodology for interpreting airquality measurements from urban streets using CFD modelling. Atmos. Environ.45, 5230e5239.

Soulhac, L., Salizzoni, P., Mejean, P., Perkins, R.J., 2013. Parametric laws to modelurban pollutant dispersion with a street network approach. Atmos. Environ. 67,229e241.

Stern, R., Yamartino, R.J., 2001. Development and rst evaluation of MICRO-CALGRID: a 3-D urban-canopy-scale photochemical model. Atmos. Environ.35, S149eS165.

Vardoulakis, S., Fisher, B.E.A., Pericleous, K., Gonzalez-Flesca, N., 2003. Modelling airquality in street canyons: a review. Atmos. Environ. 37, 155e182.

Venegas, L.E., Mazzeo, N.A., 2006. Modelling of urban background pollution inBuenos Aires city (Argentina). Environ. Model. Softw. 21, 577e586.

Venegas, L.E., Mazzeo, N.A., Pineda Rojas, A.L., 2011. Evaluation of an emission in-ventory and air pollution in the Metropolitan area of Buenos Aires. In:Popovic, D. (Ed.), Air Quality d Models and Applications. InTech Pub, Croatia,pp. 261e288.

Wang, T., Xie, S., 2009. Assessment of trafc-related air pollution in the urbanstreets before and during the 2008 Beijing Olympic games trafc control period.Atmos. Environ. 43, 5682e5690.

Yamartino, R.J., Wiegand, G., 1986. Development and evaluation of simple modelsfor ow, turbulence and pollutant concentration elds within an urban streetcanyon. Atmos. Environ. 20, 2137e2156.

Zhang, Y.-W., Gu, Z.-L., Cheng, Y., Lee, S-Ch, 2011. Effect of real-time boundary windconditions on the air ow and pollutant dispersion in an urban street canyon-Large eddy simulations. Atmos. Environ. 45, 3352e3359.

L.E. Venegas et al. / Atmospheric Environment 87 (2014) 77e8686

A simple model for calculating air pollution within street canyons1 Introduction2 The SEUS model2.1 Model description2.2 Sensitivity analysis of the basic expression of SEUS2.3 Chemical processes in the street canyon2.4 Study of the behaviour of SEUS with ambient wind

3 Results and discussion3.1 Site and data3.2 Comparison of model results with observations3.3 Comparison between SEUS and OSPM performances in Crdoba Ave

4 ConclusionsAcknowledgementsReferences